paper-buf

• LLM safety
  • Scalable agent alignment via reward modeling: a research direction
    • NOTES
  • Measuring progress on scalable oversight for large language models
  • Model stealing for any low-rank language model
  • Lazy safety alignment for large language models against harmful fine-tuning
  • Reducing the scope of language models with circuit breakers
• Knowledge updating
  • MiniRAG: Towards extremely simple retrieval-augmented generation
  • Cluster-norm for unsupervised probing of knowledge
• Others
  • Improving instruction-following in language models through activation steering
  • Musing about future of search: A return to the past?
  • Machine unlearning
  • Adversarial examples might be avoidable: The role of data concentration in adversarial robustness

Killed by LLM

• A website that archives the benchmarks LLM has saturated on.

How I use Claude

• A blogpost by a user outside LLM developers and AI researchers on how he use Claude in his daily work/life. Very interesting.